Predictive Modeling of Cognitive Decline in Alzheimer’s Disease via Multi-Modal Digital Twins: Algorithmic  Limitations, Explainable AI, and Healthcare  Management Perspectives

Elcin Huseyn

Abstract. The heterogeneous progression of Alzheimer’s disease (AD) continuously challenges the boundaries of traditional clinical neurology, particularly in early diagnosis and the personalized administration of disease-modifying therapies. In recent years, the paradigm of the “digital twin” a dynamic, data-driven in silico replica of a patient’s neurobiological profile, has emerged as a theoretical solution to this heterogeneity. Despite its growing prominence in computational neuroscience, the current literature reveals significant methodological fragmentation regarding data fusion strategies and algorithmic architectures. This critical review synthesizes recent advancements (2020–2025) in multi-modal digital twin models aimed at predicting cognitive attrition in AD. We critically analyze the limitations of unimodal biomarker reliance and the polarizing debate between ensemble learning methods and Deep Neural Networks (DNNs). Our synthesis indicates that while Deep Learning achieves superior diagnostic accuracy in spatial topography, its inherent algorithmic opacity (“black box” phenomenon) severely undermines clinical utility and trust. Furthermore, we evaluate the implementation of these models through the lens of healthcare management, emphasizing how predictive simulations can optimize resource allocation and transform health policy. We conclude that the future of neurodegenerative care relies not merely on algorithmic complexity, but on the integration of Explainable Artificial Intelligence (XAI) to support transparent, multi-modal clinical decision-making.

Keywords: digital twin, Alzheimer’s disease, cognitive decline, computational neuroscience, Explainable AI, multi-modal data fusion, healthcare management